-
Notifications
You must be signed in to change notification settings - Fork 109
/
CHANGES
566 lines (390 loc) · 23.2 KB
/
CHANGES
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
* New rebalance/drain code. See the docs/wiki for how to use.
***NOTE*** Old drain/rebalance code is disabled. Setting a device
into drain mode no longer does what you think it would.
2010-08-13: Release version 2.37
* Make HTTP DELETE's work reliably
* Fix obscure crashes related to "unknown devices"
* make it slightly harder to crash trackers with bad slave DB settings
(Tomas Doran)
* make tracker -> storage node connection timeout configurable.
was hardcoded to 2 seconds. (Tomas Doran)
* update_class command (Tomas Doran)
2010-04-16: Release version 2.36
* Bump required version of perlbal for mogstored.
* Critical bug fix for creating files.
If you have many dead devices, poor filter ordering could cause
MogileFS to not consider any alive devices for new files.
* Actually default to BIGINT under MySQL...
Bug would always initialize new setups to INT.
2010-04-02: Release version 2.35
* Nuke ancient mogilefs 1.1 replication code.
no_unreachable_tracking and old_repl_compat variables are gone.
(dormando)
* Support modifying class replpolicy strings (requires new
mogadm/MogileFS::Client) (dormando)
* Print fidid and devid on errors when fsck fails to fix something
(t0m)
* Allow list_keys to work with a prefix of '0' (dormando)
* Print timestamp when logging to STDOUT (t0m)
* Sort by free space and weight fewer devices to avoid CPU churn
(dormando)
* Fix deadlock-related crash bugs in db queue fetch routines.
* Fix dependencies in MogileFS::Server (t0m)
* Fix bug in list_keys command that prevented searches with a
'0' prefix. (Ask Bjørn Hansen)
2009-12-05: Release version 2.34
* Fix bug in new copycount reduction code. If it was more than off by
one it wouldn't reduce the count properly.
* Improve accept loop. A reasonably busy tracker could start dropping
new client connections too easily.
2009-11-24: Release version 2.33
* 'mogadm fsck status' will now be more accurate.
* Support non-GNU df (paul armstrong)
* Remove old !replication command. It's in `mogadm stats` and tends
to kill the parent process.
* Force refreshing internal monitoring status every 5 minutes. Help
work around a potential bug where state changes are missed.
* Use new delete queue for tempfile culling
* Calculate free space from available space, not total space
(todd lipcon)
* FSCK code has new tunables to control how aggressive it runs:
queue_size_for_fsck (how many fids to keep in file_to_queue table)
queue_rate_for_fsck (how many fids to put into the table per second)
the generic internal queue is now tunable:
internal_queue_limit (how many fids to pull out of a queue table at
once. old default was 500).
All are tunable by 'mogadm settings set'
(dormando)
* Replication will now attempt to handle over replication situations.
FSCK will catch over replication as a policy violation, and
replicate will attempt to fix it. (dormando)
(rarely it might need multiple fscks)
* Since FIDSizes is no longer used, remove it and remove gearman
dependency. (dormando)
* FSCK no longer rabidly fills the queue table. (dormando)
* Make storage code resilient to rare deadlocks (dormando)
* Add 'max_handles' config option to restart a DB connection if there
are too many handles to it (victori).
2009-10-18: Release version 2.32
* No changes... MFS is hard to release due to mogdeps :/
2009-10-18: Release version 2.31
* Fix create_close race condition. (pipegeek)
* Fix bug in reporting of available disk space.
* Many minor bug/documentation fixes.
* MySQL and Postgres installs now use BIGINT's for fid/length by
default. (dormando, robbat2)
* Accept 405 responses to MKCOL from the WebDAV server since it often
just means that the directory already exists. (Ask Bjørn Hansen)
* Fix showing domain and class in mogadm output when there's only the
default class in the domain (Mark Imbriaco)
* mogstored: pidfile support
* New FSCK work and error for BCNT for bad devcount values (they would
not otherwise be caught unless the triggered another fix).
2009-01-12: Release version 2.30
* fsck will now fix broken 'devcount' cols.
* schema version bump to 12, and new delete worker code to use the new
queue system. delete should be faster and more scalable.
* add server setting queue_rate_for_fsck to allow rate limiting how
fast the new distributed fsck creates jobs. the original default of
10k could utterly defeat wimpy databases.
* Support 0 byte files.
* Fix a bug where rebalance would accidentally delete copies of a fid.
* Create new internal queue management system, with new job
'JobMaster'. This greatly reduces database load and increases
throughput for workers using the new system.
* Add 'no_unreachable_tracking' config. Disables more of the old
replication code. Use this if you have no 1.x trackers.
* Add hook for alternate replication sources. Obscure, but can be
nice.
* Add replicate_order_final_choices hook, so multidc replication may
pull from the local network if one copy already exists.
* Fix subtle bugs in process management.
* Allow overriding of default classes.
2008-08-18: Release version 2.20
* One more time, and hopefully never again. No changes. META.yml
fixes.
2008-08-18: Release version 2.19
* No changes except for META.yml fixes.
_should_ be able to upload releases without owning half of CPAN.
2008-08-16: Release version 2.18
* Add repl_use_get_port option (Brian Lynch)
* Remove tempfile data and rows on create_close (Chaos Wang)
* Don't delete files accidentally during tempfile table cleanup
* Specify username to run the mogstored daemon (Jonathan Share)
* Many spelling errors corrected.
* PostGres locking routines did not include the hostname for the lock
release.
* RFC 2518 says we "should" use a trailing slash when calling
MKCOL. Some servers (nginx) appears to require it. (Spotted by
Timu Eren).
* Add server option 'rebalance_ignore_missing' in mogilefsd config
that allows the rebalance to ignore files that are already missing
when it tries to delete them.
* Fix MultipleHosts replication policy to no longer return desperate
devices after no longer being desperate, instead returning a state
which will queue for retries in the future till it can be truly
happy. This prevents files from being replicated to all writable
devs when there are two few writable hosts to satisfy the policy.
* Add --chmod-mountpoints option to mogautomount, in an attempt to
prevent mogstored machines from accidentally writing to their under-
lying filesystem.
* New optional argument 'startpos' to fsck_reset.
* Use correct per-Store port, defaulting to UNIX sockets on localhost.
* add --dbport option to mogdbsetup (Ask Bjoern Hansen)
* Fix state_change() bug that could make file_to_delete_later
grow unnecessarily (Radu Greab)
* Add Mogstored::HTTPServer::None to allow using mogstored
with unmanaged backend DAV servers.
* Add experimental cmd_edit_file. Client requires DAV MOVE and
partial PUT on the store to use it.
* make mogstored and mogilefsd use the now-bundled dependency
Perl modules as a last resort, unless MOGILE_NO_BUILTIN_DEPS
environment variable is true. this makes it easier for people
to use mogilefs, without dependency hell. distros/packagers
are expected to ignore (or scoff at) this "feature". but
easier for people working from tarball/svn releases. bundled
dependencies will be updated over time, as needed.
* Bring the SQLite store up to date
* Increase the timeouts in the test suite
* In the test suite, also search for mogadm in /usr/local/bin/ and
/usr/local/sbin (spotted by Dean Wilson)
* SCHEMA VERSION 10: 'replpolicy' column on 'class' table.
it's safe to --ignore-schema-version and run this mogilefsd
against an older schema. new column enables per-class
pluggable replication policies. it was always
abstract, but not easy to plugin your own alternatives.
see doc/pluggable-replication-policies.txt
* add start of a MogileFS::ReplicationPolicy::Union implementation
* fix crash in queryworker's create_open command, when a device
has its directory made on a mogile storage node (the host of
which is being monitored by the 'monitor' process), but the
device isn't registered in the devices table yet.
2007-07-03: Release version 2.17
* get_paths command now takes a pathcount option, defaulting to
2, instead of having a hardcoded constant of 2. This allows
requesting more than 2 paths to a file.
* make fsck log over-replicated files as policy violations. it
won't delete over-replicated files, though, just log
* Document how to test MogileFS under doc/testing.txt. You really
should read this if you want to use MogileFS to store your data.
* Add a safety check to ensure that Postgresql-8.2 or newer is used.
* Update mogdbsetup usage, and ensure it chooses the correct admin
username automatically.
* Document the --type option to mogdbsetup.
* Introduce the PostGres Store implementation. The locking system might
still have issues, beware of stale locks! Initial prototype by Radu
Greab <[email protected]>, and developed heavily by Robin H. Johnson
* start of support to make mogstored wrap Apache + web_dav
* Remove duplicated [fsck] in log output of Fsck worker.
* Handle HiRes gettimeofday correctly to get useful results. Arrays are
important. This gives us correct timing output.
* Handle input to log() correctly. Based on the implementation of
Sys::Syslog::syslog(), produces no ugly warnings, and makes sure that
printf stuff works nicely.
* The unix_timestamp function belongs to the Store, not other classes.
Need to add test cases for process_deletes and stats.
* Using NULL during an INSERT for the AUTO_INCREMENT column does not
work very well outside of MySQL. Pass the correct number of arguments
to the INSERT instead.
* Clean up all raw usages of UNIX_TIMESTAMP() function, in preparation
for database independence. If you need the unix timestamp, include the output
of $self->unix_timestamp directly into your query.
* Alter $0 so that the scripts show up as '$SCRIPTNAME' in the process
listing instead of '/usr/bin/perl $SCRIPTNAME'. This makes init.d
scripts easier to manage.
* make mogstored's lighttpd config use mod_status for "/", so
mogadm doesn't think it's dead when it previously saw a 404
(Justin Huff <[email protected]>)
* remove Linux::AIO support, now that Perlbal has removed it as well,
and because Linux::AIO support (when Perlbal then chooses
IO::AIO) was hanging mogstored on shutdown. so also add a
mogstored shutdown test (after a full daemonization of it)
* make Mgd::validate_dbh() re-check/ping slave dbhs as well.
(problem mentioned by Steven Shou <[email protected]>)
* support drive identifiers above sdz (Arjan <[email protected]>)
* make mogstored's iostat child process properly shutdown,
even when iostat binary isn't installed. (previously
it was stuck in a loop every 10 seconds, hoping for
iostat to be installed, but never respecting parent
process' death)
* Change delete worker to use new DeviceStates interface.
* Fix get_paths to honor new device state of 'drain' via new
DeviceStates interface.
* use significantly less CPU when re-broadcasting around
iostat device utilization% to child processes. (future:
don't rebroadcast at all and use shared memory)
* slightly faster fsck (do less syscalls)
* efficiency: don't broadcast dup state information to children
(track in parent what children know already)
* if using Gearman::Server >= 1.09, respawn the fidsizes
worker if it dies.
2007-05-09: Release version 2.16
* BUG FIX from 2.15: partially written but never closed
tempfiles weren't being cleaned up properly.
* start to work better when compiled as a stand-alone
PAR all-in-one script/archive
2007-05-07: Release version 2.15 ("fsck/drain/rebalance")
* minor bug fixes and enhancements for MultipleHosts replication
policy. bunch of corner cases now checked with new tests.
now easy to write more replication policy tests in future.
* replication policies can now return "desperate" requests,
signalling that a replication reassessment should be
enqueued for the future, to see if things could be improved.
(that part's not currently implemented, but the real feature
and motivation is that the rebalancer now won't delete a
DevFID if it results in a desperate move, only an ideal move. )
* replication policies now can optionally return a new return
value of the (new) type MogileFS::ReplicationRequest, which
has pretty accessor names, can suggest multiple places,
can indicate non-ideal emergency replication decisions.
old plugins' return values will be transparently upgraded
to the equivalent new return value objects.
* adding new device state: "drain". it's a hybrid of "dead"
(in that files are migrated off it) and "readonly", it that
it still serves traffic... it just doesn't get new files.
this also introduces the new object-oriented DeviceState class,
and device_state($name) utility function to get the DeviceState
singleton by name
* internal code cleanup. notably, kill the old & nasty legacy
'find_deviceid' function which was ridiculously long and hairy.
the two callers are now more readable with sorts/greps/etc.
* make mogstored's devN/usage writing process (DiskUsage) be less
racy with the mogilefsd monitoring code... don't open file for
write... open read/write, then in one write system call, write
the entire file, with newline padding at end to cover old data,
then truncate it if necessary. should remove harmless (but scary)
error messages previously reported by the mogilefsd monitor
about zero-length usage files.
* new protocol commands to list/set (certain) server settings,
with value sanity checking (see MogileFS::Config for which
are settable, and with what values). needed for "enable_rebalance".
was partially enabled before for slave settings. also needed
for memcached support before, which was never possible to
set with mogadm, only with db tweaking.
* make mogilefsd fsck use new mogstored fid_sizes command, to
do bulk stats. speeds up fscks a ton.
* be robust against system clocks that go backwards between
gettimeofday calls:
http://lists.danga.com/pipermail/mogilefs/2007-April/000897.html
* Put gearman server in mogstored process, add worker
'mogstored-fidsizes' which runs as subprocess of mogstored. Add
side-channel command 'fid_sizes' which allows us to quickly enumerate
and get sizes for files across entire devices on a storage node.
* remove all code like $state eq "readonly", $state=~ /^dead|down$/
and instead convert it into specific questions on policy/traits
of given state, like $dev->should_put_new_files_on, or
$dev->should_drain_files_off. see MogileFS::DeviceState,
objects of which are accessed via $dev->dstate, or new
MogileFS::Util device_state($name) wrapper.
* start of rebalance support. (where replication workers, in their
idle time, can rearrange files to even out disk space and/or IO
activity on storage nodes... policy isn't hard-coded, and is
in fact currently random)
* lighttpd support in both mogilefsd and mogstored. passes test
suite with environment MOGSTORED_SERVER_TYPE=lighttpd set now.
* abstract out the HTTP server support in mogstored, so
mogstored isn't just a perlbal wrapper, but an anything
wrapper. (in particular, lighttpd and apache) mogstored still
exists for all its other misc admin/monitoring functions,
but can then manage/configure apache/lighttpd child process(es).
so far they're just stubbed out.
* split mogstored into separate files per class, rather than one
large script.
2007-04-24: Release version 2.10
* bug from earlier refactoring: changing some 'next' to 'return'
in MogileFS::Device's create_directory method. (thanks to
Justin Huff <[email protected]>)
* make sure mogdbsetup doesn't upgrade an existing table (with
potentially lot of data) from MyISAM to InnoDB. just warn.
only new tables should become InnoDB on creation. also,
detect InnoDB even on old MySQL 4.0.x, not just 4.1/5.0.
* make fsck faster by reducing number of db queries (both
reads and writes), though more unrelated fastness remains to
be done.
* make fsck_status faster (keep summary stats of evcode counts,
rather than doing a full GROUP BY query all the time)
2007-04-20: Release version 2.09
* filesystem checker (Fsck worker). off by default, unless
it's enabled via protocol. Control/monitor it with mogadm
in MogileFS-Utils 2.08, which needs MogileFS::Client 1.06.
* replication speed-up on configurations with only 1 host.
(replication would randomly have no suggestions before, when
there was only 1 target host, so replication would be rescheduled
until the future, when random returned a different device on
that host)
2007-04-17: Release version 2.08
* (no runtime-visible changes from 2.07)
* mogilefsd: remove incomplete "Checker" worker, replace with
incomplete "Fsck" worker.
* test cleanup/robustness improvements. properly skips testing when
testing deps not found. but searches harder for them now, too.
tests run under "make disttest".
* packaging cleanups, released using ShipIt.
2007-03/04:
* mogilefsd: Split Plugin::MetaData and Plugin::FilePaths into their own distributions.
* mogilefsd: Fix mogdbsetup to not ask empty questions.
* mogilefsd: command line/config file support for listening on specific IPs, also
support for listening on multiple IPs/ports
2007-03-19:
* mogstored: use XS HTTPHeaders by default, unless disabled with environment variable
* mogilefsd: memcache support for get_paths (see doc/memcache-support.txt)
2007-03-16:
* mogstored: make child processes be exec'ed (as as their own scripts, not inheriting
Perlbal, Linux::AIO/IO::AIO, etc baggage). so they don't own listening
sockets, pipes to child workers, signal handlers, etc. just a lot
cleaner now, and SIGTERM, SIGINT properly shut everything down now,
both in foreground and as daemon
2007-03-15:
* mogstored: this changelog is totally neglected. :/ see svn log.
* mogstored: make perlbal's "shutdown graceful" work, by not doing
sidechannel accepting ourselves (which never hung up),
but subclass Perlbal's TCPListener class, which does it
all correctly.
2006-12/2007-01: (2.00_05)
* mogilefsd: everything converted to OO API internally. old internal APIs removed.
* mogilefsd: be database-agnostic. Currently only works with MySQL (and
SQLite as a joke/demo), but Pg/Oracle coming in the future.
* mogilefsd: built-in I/O activity monitoring between mogstored/mogilefsd Monitor
now.
* mogilefsd: much improved internal communication between worker processes
* mogilefsd: start of better docs
2006-12-28
* mogilefsd: pidfile support (Radu Greab <[email protected]>)
2006-12-28
2006-12-27
* mogilefsd: remove all old internal APIs, update everything to use new OO APIs,
expanding OO APIs where necessary (lots of places). so much more
beautifully hackable now.
* mogstored: kill child processes (diskusage and iostat) when being killed.
2006-11-05:
* mogstored: set aio threads after daemonizing. needed by Linux::AIO/IO::AIO, it seems?
2006-09-29
* mogilefsd: remove non-HTTP mode ("NFS mode") from the server
2006-09-23:
* mogstored: refuse to start up with missing Linux::AIO/IO::AIO or broken Linux::AIO
2006-09-04:
* mogstored: set aio_threads as a function of number of disks
2006-07 -> 2006-09
* mogilefsd: major refactoring/cleanup/new features ("MogileFS 2.0")
2006-03-01 (Brad)
* mogilefsd: fix dataloss bug when the 'tempfile' table is InnoDB and the
server is restarted while the tempfile table is empty.
2006-02-28 (Jr)
* mogilefsd: allow create_open to take optional fid parameter which enables you to
dictate which fid to use internally and to not rely on MySQL's auto_increment
2006-01-10 (Brad)
* mogilefsd: make Mgd::get_device_summary() always return all devices, regardless
of status.
* mogilefsd: update all callers of get_device_summary to filter on the device
status, if they need to. (provided more clarity into what
callers' intentions were...)
* mogilefsd: add a new device state 'readonly' which means: it can be read from
from clients, it can be replicated from, but no new files can go
on it or be replicated onto it. and it also reaps, meaning the
device is fucked. the filesystem probably ate itself.
2005-08-10:
* mogstored: write out usage files for devices, even if they're not on their own mount point.
also explicitly include the '-k' flag to df, so we know they're KiB.
2005-08-07:
* mogstored: new config syntax and command line options